In recent decades the importance of disturbances on the forest carbon and water cycles have been recognized as well as the effects of climate change in modulating these (Dale et al., 2001; Kurz et al., 2008a; Reichstein et al., 2013; Gatti et al., 2014). Future predictions of forest recovery and health depend on an understanding of current mechanisms of mortality and an understanding of forest structure, function, and underlying mechanisms of species compositional dynamics under disturbance regimes (Seidl et al., 2011a). To date, most models do not take into account physiological changes, trade-offs in response to multiple forest disturbances (physical and biogenic), feedback mechanisms between nutrients and forest species, or potential species shifts (Dietze et al., 2011, 2013; Medvigy et al., 2012; Richardson et al., 2012). In addition, mechanisms of mortality are not well understood and thus not incorporated into models (McDowell et al., 2008, 2011). Ecosystem response to extreme climate events such as drought can result in increases in defoliation, fire or wind-throw (Ayres and Lombardero, 2000; Dale et al., 2001; Reichstein et al., 2013) and a decrease in transpirable soil water content (Klein et al., 2014). Forest functioning and species composition will likely be altered by re-occurring droughts, insect infestations and windthrow, while the changes in energy partitioning will likely have impacts for regional climate in forest ecosystems (Roy and Avissar, 2002). This, in turn, could increase fire risk (Smithwick et al., 2009; Seidl et al., 2011b; Stephens et al., 2013). Conversely, climate extremes can have delayed feedback impacts on soil water content, and thus, ecosystem function (Reichstein et al., 2013). Additionally, species may vary in their responses to such occurrences (Schäfer, 2011). Therefore, specific ecosystem responses are not well known and are difficult to model due to a lag in response (Reichstein et al., 2013).

Clearly, in order to build predictive models, the processes need to be captured on the leaf and/or canopy scale. While canopy net assimilation scaled via sapflux (see Schäfer et al., 2010) vis a vis gross ecosystem production measured with eddy covariance (see analysis in Amiro et al., 2010) show overall reduced carbon uptake after insect attack, the process on the leaf-level shows compensatory responses such as higher photosynthetic activity (Heichel and Turner, 1983; Vanderklein and Reich, 1999) or water use per unit leaf area (Meinzer and Grantz, 1991; Schäfer, 2011); even under drought conditions (Hawkes and Jon, 2001). However, nutrient removal via defoliators could reduce photosynthetic capacity over time, thus effectively hindering recovery (Krause and Raffa, 1996). Therefore, the overall reduction at the canopy scale is mediated through leaf-level compensations rather than just a function of reduced leaf area as it is implemented in models (see Medvigy et al., 2012). In contrast, prescribed fires have only short-term effects on overstory trees or the ecosystem at large (Clark et al., 2012; Renninger et al., 2013), given that they mainly affect understory shrubs and forest floor fuel loading (Boerner, 1981; Boerner et al., 1988). Any effect on overstory trees or ecosystem scale carbon and water cycling are transient (Clark et al., 2012; Renninger et al., 2013). Wildfires, however, have a devastating effect on the water and carbon balance of forests, as they often are stand replacing or largely more destructive to overstory trees (Hurteau and North, 2008, 2009; Hurteau et al., 2008, 2011; Wiedinmyer and Hurteau, 2010). Furthermore, these wildfires as disturbance regimes can potentially play a huge role in forest health and structure (Heinselman, 1973). Depending on the burn regime, fires can lead to both horizontal and vertical structural changes by altering canopy gaps, species composition, and tree densities, which can then subsequently alter competitive relationships (Heinselman, 1978; Boerner et al., 1988). It has also been found that fire can have a direct impact on physical and chemical properties of the soil (Granged et al., 2011), which could potentially lead to altered physiological responses of the overstory trees, further affecting the carbon and water budgets. Thus, insight into hydrodynamics (Lopushinsky and Klock, 1980; Bohrer et al., 2005; Thomsen et al., 2013), nutrient limitation (Lovett and Tobiessen, 1993; Krause and Raffa, 1996; Vanderklein and Reich, 2000) or enhancement of photosynthetic capacity (Heichel and Turner, 1983; Haukioja et al., 1985; Hodgkinson, 1992; Vanderklein and Reich, 1999) in response to disturbances such as drought, insect defoliation, and fire would help build better predictive models to assess forest structure, function, and species compositional shifts under disturbance regimes. This will help improve predictions of water and carbon cycling of forest ecosystems.

Capitalizing on a long-term data collection effort in a xeric forest of the Atlantic Coastal Plain, the New Jersey Pine Barrens, insights into drought and prescribed fire (as a physical forcing agent) and insect defoliation (as a biogenic forcing agent) plant responses have improved our understanding of plant compensatory responses, potential mortality agents and species compositional shifts, thus enhancing predictions of water and carbon cycling of forests (Schäfer et al., 2010, 2013; Schäfer, 2011; Clark et al., 2012; Medvigy et al., 2012, 2013; Renninger et al., 2014b). It is important to note differences in physical disturbances, such as fire and windthrow that are non-species specific and biogenic disturbances, such as defoliators or phloem feeders that are species specific and thus have a larger impact on forest dynamics and species compositional changes. Here, we provide a synthesis and insights of the effects of physical and biogenic disturbance to water and carbon cycling in upland forests of the New Jersey Pine Barrens.

Measurements of transpiration, ecophysiological parameters, biometric variables and eddy covariance measurements in an oak/pine ecosystem in the Atlantic Coastal Plain (New Jersey Pinelands) showed a relative conservatism of water use (Clark et al., 2012) on an ecosystem level, but longer lasting effects on carbon balance after insect defoliation. While post-defoliation (2012) transpiration and evapotranspiration are similar to pre-defoliation levels (2006), post-defoliation carbon fluxes have not returned to pre-disturbance levels after 5 years of recovery due to a 25% reduction in basal area following tree mortality (Schäfer et al., 2013). Defoliation frequency also affects recovery, with modeled carbon fluxes under various defoliation scenarios showing pronounced reduction in productivity under frequent defoliation, but no effect if defoliation occurs at a rate of >15 years (Medvigy et al., 2012).

Despite a relatively consistent seasonal water use through various disturbances, defoliation and drought affected water use differently. For example, canopy transpiration (E_C) after defoliation and subsequent re-sprouting, was reduced by 25% compared to pre-defoliation levels, even though only half of the leaf area was replaced. However under severe drought conditions in 2006 and 2010, E_C was only reduced by 8 and 18% respectively (Table 1, Schäfer et al., 2013). Therefore, prolonged drought had a lesser effect on E_C than reduced foliage or episodic defoliation, suggesting these trees have access to deeper soil moisture. These data also suggest that defoliation may make trees more sensitive to drought over time as evidenced by the higher reduction of E_C during a 2010 drought period (post-defoliation) compared to a 2006 drought (pre-defoliation) (Schäfer et al., 2013).

Differential physiological responses of the various oak species as well as pitch pine may also create a species shift in an ecosystem that is also prone to fire (Table 2). In this ecosystem, Quercus prinus showed consistently lower stomatal conductance, photosynthesis and maximum carboxylation rate compared to Quercus velutina, however both oak species displayed similar water and nutrient use efficiencies (Renninger et al., 2014b). Likewise, Pinus rigida, a predominant species in the Pinelands, showed comparable water– and nutrient use efficiencies to the oak species investigated signifying similar strategies in this ecosystem with respect to their efficiencies. However, Q. velutina had higher mortality rates than Q. prinus suggesting a possible shift in oak species with more frequent defoliation events (Schäfer, 2011). Likewise, P. rigida may be released from competition if more oaks species face mortality due to gypsy moth defoliation occurrences (Medvigy et al., 2012).

Prescribed fire in this ecosystem had a short-term effect on leaf-level and canopy-level stomatal responses (Figure 2, Table 2). Leaf-level stomatal responses remained unchanged in relation to the coupled control site for P. rigida, directly following the fire (Figure 2). Comparing the pine site with the upland oak/pine site, increased water use by overstory pines was observed, while at the oak/pine site, the fire decreased stomatal conductance the summer after the fire. Therefore, there could be differing effects depending on stand type with the pine-dominated stand being positively affected by fire and the oak-dominated site being negatively affected. For example, pre-fire canopy stomatal conductance (G_c) at the pine fire site was significantly higher than the control site (P = 0.01). However, following the fire, the control and fire site exhibited no statistical difference (P = 0.3). In this forest ecosystem, prescribed fire, therefore, has little effect on the leaf-level physiological responses of overstory pitch pines (Figure 2, Table 2). While some initial trends were noticed in discrete cases, these responses did not hold true across the two prescribed fires sites and thus show differential responses across different stands with different species composition. However, a common trend that did seem to develop was a transient response to a prescribed fire. In some cases, such as carboxylation efficiency and maximum assimilation with respect to increase in carbon dioxide concentration, there was an initial increase following the fire, which subsided by the first or second summer after the fire (Renninger et al., 2013). Another common trend was a delayed response in which physiological differences from late winter/early spring prescribed fires were not seen until the summer growing season. For example, no initial increase in either photosynthetic capacity (V_cmax) or intrinsic water use efficiency was noted in the weeks post fire, indicating a lag of response until new needles are produced capitalizing on release of nitrogen post-fire. However, a large increase in these two parameters was measured by the summer growing season. These trends suggest that prescribed fires affect stands differently depending possibly on fire intensity, fuel loading and species composition.

The major knowledge gap to understand and thus model disturbance, recovery and resilience are that most terrestrial or demographic vegetation models [such as BIOME-BGC (Running and Gower, 1991), ED2 (Medvigy et al., 2009) etc.] do not take into account disturbances such as fire, insect defoliation, hurricane or snow load disturbance (McCarthy et al., 2006) and their physiological impacts. In addition, physiological responses in general and parameterization thereof are ill defined in models (Rogers, 2014). Defoliation, for example, has only been implemented through leaf area reduction, but does not take into account compensatory photosynthetic responses (Medvigy et al., 2012). Since, often, photosynthetic capacity (Rogers, 2014) or stomatal conductance are ill-defined in these models [meteorological driven models such as the Ball-Berry Model, (Ball et al., 1987; Medvigy et al., 2013)], it is difficult to incorporate changes due to disturbances that have physiological effects (see Table 1) that are known to be important (Thornton et al., 2002; Rogers, 2014). Likewise, species compositional changes are unknown after disturbance, and at the ecosystem level, responses may be delayed and cannot be measured until years after a disturbance or extreme climatic event (Boerner, 1981; Runkle, 1981, 2000; Reichstein et al., 2013). In addition, recent reports suggest that the Southern Pine Bark Beetle will invade the NJ Pine Barrens potentially increasing mortality to pine species (Gillis, 2013). Therefore, future species composition in this forest depends on a range of insect disturbances, which are driven by climate change making the species dominance outcome unclear. Generally, species composition after physical disturbance changes very little, as the forest gaps are filled with species already present (Runkle, 1981, 1982, 1984; Frelich and Reich, 1999). However, as biogenic disturbances are more species specific, the dynamics are less clear (Kurz et al., 2008b; Seidl et al., 2011a).

Measured plant compensatory responses can confound ecosystem level responses to disturbances, particularly if they lead to delayed responses (Sala et al., 2010). In addition, release from competition can confound or enhance plant physiological responses to disturbances (Wickman, 1980; Runkle, 1981; Runkle and Yetter, 1987; Tilman et al., 1997; Frelich and Reich, 1999; Vanderklein and Reich, 1999). Plant compensatory responses have been well documented and are similar to our findings (Reich et al., 1993; Vanderklein and Reich, 1999; Clinton et al., 2011; Schäfer, 2011), however the incorporation into models is still lacking. Thus, devising strategies to manage forests are yet hampered by this limitation (Seidl et al., 2011a). In addition, the effects of disturbances are not necessarily perceived in a matter of years but rather decades (Baker, 1941) with potentially compounding effects (Stevens and Beckage, 2009; Gaylord et al., 2013; Schäfer et al., 2013). Even if management of disturbances is implemented, such as species compositional changes, the outcome may take decades to take effect (Seidl et al., 2008, 2009).

In the case of prescribed fire (Table 2, Figure 2), the pine site has been positively effected by the prescribed fire, because of a thicker duff layer, surface roots and microbes may have not been damaged (Boerner, 1981; Boerner et al., 1988). However, at the oak/pine site, a thinner duff layer may have led to higher temperature effects at the soil surface, thus potentially damaging some of the surface roots and microbes (Varner et al., 2009). Therefore, plant functional types play an important role in the structure and function of these forests. However, since the responses are short-term and transient (Clinton et al., 2011), the long-term impact on the carbon and water cycling in these forest ecosystems are likely to be small.

While there may be a general framework to assess tree mortality in response to drought and insects (McDowell et al., 2011), ecosystem responses to drought and insect attack (folivory or phloem feeding) are contingent on individual tree trade-offs, which are themselves contingent on tree ontogeny. Barbeta et al. (2003) found that larger trees survived a long-term drought treatment better than smaller trees, presumably because as the smaller trees died, they freed up soil moisture for the larger trees, which may have a combination of deeper root systems and a higher water storage capacity. However, the mortality of the smaller trees must be the result of trade-offs between growth and ability to respond to drought. If smaller trees have higher root to shoot ratios (Kearsley and Whitham, 1989; Boege and Marquis, 2005), yet are more susceptible to drought, then carbon stores and the ability to utilize those stores must be more important for survival than drought resistance per se. On the other hand, higher resistance to drought may result in lower maximum assimilation and water use efficiencies (Limousin et al., 2010). Interestingly, this may not be the case in xeric environments, such as the New Jersey Pine Barrens in the Atlantic Coastal Plain investigated here (Schäfer, 2011; Renninger et al., 2014b). Furthermore, larger trees and trees growing in arid regions have larger non-structural carbohydrate pools (Sala et al., 2010) suggesting that they should be less vulnerable to mortality as a result of carbon starvation. Likewise, results from defoliation research using seedlings show that trees may retain a minimum amount of carbon regardless of defoliation intensity (Chapin et al., 1990; Reich et al., 1993; Vanderklein and Reich, 1999). Thus, a distinction needs to be made between total carbon pools and available carbon pools (McDowell et al., 2011).

The interactions and possible trade-offs between tree responses to insect attack and drought are unknown (Agrawal, 2007; Jactel et al., 2012), whereby carbon used for defense against insects cannot be used for repair (i.e., of cavitation induced by drought). Plants may also reduce their carbon demand by reducing respiration rates and/or shedding plant parts in response to drought (Sala et al., 2010). Functionally, shedding plant parts should be similar to defoliation depending on what is shed. On the other hand, a possible trade-off for increased drought resistance could be higher susceptibility to insect attacks (Mattson and Haack, 1987). However, as was also shown here in an upland oak/pine forest in NJPB, Q. prinus not only withstood drought better, but also sustained less mortality after gypsy moth disturbance compared to Q. velutina (Schäfer, 2011). The differences in mortality may be due to different resource use strategies, whereby Q. velutina was shown to have higher photosynthetic capacity and nitrogen (N) per unit leaf area, thus was more vulnerable to N removal through insects (Renninger et al., 2014b). Thus, as has been shown before, different species respond differently to drought (Schäfer, 2011; Wu et al., 2011; Renninger et al., 2014b) and may adapt over time to it (Wu et al., 2011) or may become more susceptible to drought over time (Hacke et al., 2001; Anderegg et al., 2013). However, ecosystem function depends not only on biotic or abiotic factors but also sociological and economic factors. The New Jersey Pine Barrens are managed forests in relatively close proximity to large, urban centers, thus how they are managed has consequences for ecosystem processes that can also affect carbon and water dynamics as was shown here with regard to prescribed fires. However, adaptive management practices may take decades to have a perceptible impact (Seidl et al., 2009), thus a forward leaping approach is needed that allows proper management decisions to be made since corrective measures will be difficult (Seidl et al., 2008).

Forest management practices have to be persistent in order to recover forest health (Seidl et al., 2008; Gormley et al., 2012), but also need to be able to address forest mortality, and thus loss in carbon sequestration potential (McCarthy et al., 2006). Research will need to improve our understanding of a) species responses to a particular disturbance, b) mechanisms leading to mortality and c) how to include this mechanistic understanding into models that, in turn, will help to predict future changes and responses of forests. As this study suggests, forest response to disturbance varies from the leaf to ecosystem level as well as species level and thus, these differential responses interplay to determine the fate of forest structure and functioning.

Forest functioning will likely be altered by re-occurring droughts, gypsy moth defoliation and windthrow of already weakened trees. However, prescribed fire has only transient responses to the carbon and water balance in this ecosystem. In this forest ecosystem, precipitation variations exerted an overriding effect on the hydrological budget compared to biological changes in this forest, thus it is likely that climate change will cause more changes to the groundwater table and therefore water supply to regional populations. However, changes in energy partitioning due to canopy gaps after mortality will likely have impacts for regional climate in forest ecosystems. Also, in a study on snags and coarse woody debris, carbon pools that quadrupled after gypsy moth-drought mortality suggests that, in a back of the envelope calculation, it will take at least 18 years before current dead wood will have respired, making the carbon balance in this forest uncertain (Renninger et al., 2014a). Thus, while the water balance in this forest ecosystem seems to recover faster within this ecosystem (Clark et al., 2012), the carbon balance has still not recovered to pre-defoliation levels. However, prescribed fire has only transient responses to the carbon and water balance in this ecosystem. Gaining a better understanding and developing a mechanistic underpinning of these responses and incorporating them into larger scale models to improve carbon and water cycle modeling is essential (Dietze et al., 2013). Of particular importance is the ability to incorporate into models the physiological responses on the leaf level and potential compensatory responses on the ecosystem level or vice versa.